Moonshot AI has launched Kimi Hosted Agents, a managed cloud service for running long-running and collaborative AI agents. The practical pitch is that companies can use a hosted agent runtime with isolated sandboxes, activity logs and controls over what agents can access, rather than building that infrastructure themselves.
Moonshot AI/Kimi Watch analysis
What happened
The service lets teams configure agents on Moonshot’s infrastructure for tasks such as monitoring, research and operations, or embed them in customer-facing products. It offers more than 140 skills and 30 plugins, and developers can attach their own tools, data sources and MCP services.
Moonshot also describes an optional Kimi Inference Hook, which lets an enterprise’s own or third-party safety service inspect task inputs and tool returns before they reach the model. The hook is currently limited to enterprise-verified customers. The launch details were reported by PanDaily, citing IT Home and Moonshot’s open-platform documentation. (Read the report)
Why it matters
A hosted runtime could spare organisations the work of maintaining the machinery around an agent, while sandbox isolation, network controls and exportable execution logs offer concrete ways to manage its activity. Those controls matter when agents are connected to internal systems, not just asked to produce another paragraph in a chat window.
The service also itemises sandbox hours, model inference and plugin calls. PanDaily reports a sandbox rate of 0.072 yuan per hour, with model and plugin charges on top. That is a useful starting point for budgeting, not a complete price for running an agent.
Our read
The interesting part is not simply that Moonshot has put agents in the cloud. It is the combination of a managed runtime with controls for permissions, logs and inspection before model calls. Those are the less glamorous parts of agent deployment, which is usually a sign they are worth asking about.
For teams considering it, the next sensible step is to establish the full cost and test how the isolation and approval controls behave with their own tools and data. A list of features is not yet an operational track record.
What to watch
- The full pricing for inference, plugins and sandbox use.
- Which controls are available to customers beyond the enterprise-verified tier.
- Whether Moonshot publishes details on reliability and performance in deployed workloads.
Discussion spark: When deploying AI agents into company systems, should teams prioritise a managed platform with built-in controls, or insist on running the infrastructure themselves?
Sources and evidence
- Moonshot AI Launches Kimi Hosted Agents as a Managed Agent-Harness Cloud Service (5 November 2026, 06:40 UTC)
- openJiuwen's Hyper-Agent Swarm Puts Persistent AI Agents Into Work Chats as Digital Employees (5 November 2026, 06:40 UTC)
- China Merchants' LionRock AI Lab Open-Sources Lion-ER-35B-A3B Embodied Reasoning Model (5 November 2026, 06:40 UTC)
not affiliated with or endorsed by Moonshot AI and Kimi